Traditionally, scheduling experts rely on their knowledge and experience to develop problem-specific heuristics that require a considerable amount of time, experience, and code effort. Through this tedious process, experts must follow a trial-and-error cycle by evaluating the generated rules in a simulation model for the problem under consideration until achieving satisfactory results. Recently, hyper-heuristic approach has emerged as a powerful technique that uses artificial intelligence to automatically design efficient heuristics for various optimization problems. Genetic programming (GP) is the most popular hyper-heuristic approach to automate the design of production scheduling heuristics. In this paper, a genetic programming framework is proposed to generate efficient dispatching rules in a dynamic job shop. The proposed framework integrates the reasoning mechanism of GP with the ability of discrete event simulation in analyzing the performance of generated rules under dynamic conditions. Afterward, the evolved heuristics are compared to human-tailored literature rules under different dynamic settings using mean flow time and mean tardiness as performance measures. The achieved results prove the ability of the proposed approach in generating superior scheduling rules rapidly, within a few hours, compared to the conventional literature rules commonly adopted in the industry.